AI Research and Knowledge Base

Why It Matters

Personalized emails outperform templates by a wide margin. But manual research — reading LinkedIn profiles, scanning company pages, finding a relevant hook — takes 10-15 minutes per prospect. Kit’s AI agent does that research automatically, then writes an email that references what it found.

The Research Pipeline

Some context reaches the AI before it makes a single tool call. The prospect’s name, company, title, source URL, your operator notes, and any prior research summary are all written straight into the agent’s prompt. If the campaign is linked to a job posting, the full posting — title, department, location, employment type, salary, description, and hiring stages — is preloaded the same way, so the AI never has to look it up.

From there the agent works a research waterfall, stopping as soon as it has enough for a specific email:

1. Fetch source URL — If the prospect has a source URL (LinkedIn, personal site, company page), the AI fetches and reads it first. This is the richest single source.
2. Search for the person — The AI searches the web for the prospect by name and company to find recent activity, blog posts, or press mentions.
3. Search for the company — The AI researches the company for context — product, industry, recent news, funding.
4. Check Kit internal data — If your account has Compensation Research, the AI reads Kit’s own hiring, salary, and tech-stack data on the company.
5. Search the knowledge base — If the campaign has knowledge keys, the AI pulls matching entries for product context, talking points, and case studies.

Steps 1, 4, and 5 only appear when they apply, so a prospect with no source URL on an account without Compensation Research runs a shorter waterfall. Recruiting campaigns can also pull up your other published roles if the prospect looks like a better fit elsewhere.

Once research is complete, the AI writes the full email — subject line and all sections — based on what it found. Each section follows the instructions and sentence limits defined in the campaign’s sequence steps.

Confidence Scoring

After researching, the AI scores its own confidence (0.0–1.0) against this rubric — it reflects what the research turned up, not how good the email is:

Score	Meaning
0.8–1.0	Found specific personal details and company data. Truly personalized email.
0.5–0.7	Found company-level info plus some personal details. Decent personalization.
0.2–0.4	Only generic industry info. The email will feel templated.
0.0–0.1	Nothing useful found. Drafted with minimal context.

On recruiting campaigns the AI is told to stay at or below 0.5 unless it can name a specific project, repository, article, or technical decision by the prospect — “your background in X” doesn’t count as personalization.

Kit then buckets that score into three bands wherever it displays research:

Band	Range	What you see
High	0.7–1.0	Green — “Multiple sources analyzed. Research is comprehensive.”
Medium	0.5–0.69	Amber — “Moderate coverage. Consider reviewing the draft and adding source URLs for better personalization.”
Low	0.0–0.49	Red — “Few sources found. Add LinkedIn or company URLs to improve research quality.”

The score shows as a color-coded percentage with that guidance line on the message review page and on the prospect’s Research tab. For anything in the amber or red band, add more prospect details (notes, source URL) or enrich your knowledge base, then re-run research.

Knowledge Base

The knowledge base stores product context, case studies, and talking points that the AI references when drafting emails. Navigate to Outreach > Knowledge Base to manage entries.

Adding Entries

Each entry has a unique key (lowercase alphanumeric with hyphens, e.g., product-overview) and a name for display. You can add content in two ways:

• Upload a file — Text or markdown files, up to 5 MB. Kit extracts the content automatically.
• Fetch from URL — Provide a public URL and Kit downloads and extracts the page content.

Linking to Campaigns

To make a knowledge entry available to a campaign’s AI agent, add its key to the Knowledge keys field in the campaign’s AI directives. The AI searches linked entries for relevant context during the research phase.

You can link multiple entries to a single campaign. Use specific, focused entries rather than one large document — the AI retrieves more relevant context from targeted entries.

Improving Draft Quality

If your drafts feel too generic or miss the mark, try these adjustments:

Action	Impact
Add prospect notes	Free-text context the AI reads before drafting — mention a recent interaction, mutual connection, or specific reason for reaching out
Include source URLs	LinkedIn profiles and company pages give the AI concrete details to personalize around
Write clear AI instructions	Tell the AI what product features to mention, what value prop to lead with, or what angle to take
Add knowledge entries	Product docs, case studies, and competitive positioning give the AI substance to reference
Use banned words	Remove buzzwords that make emails sound templated
Tighten word limits	Shorter emails (80–120 words) force the AI to be specific rather than padding with filler

Quick Checklist

• Ensure prospects have source URLs for better research
• Create knowledge base entries for your product and key talking points
• Link knowledge keys to your campaign’s AI directives
• Write clear, specific AI instructions for the campaign
• Review low-confidence drafts and add more prospect context where needed

Next Steps

• Reviewing and Sending Messages — approve, edit, or reject AI drafts
• Creating Campaigns — configure AI directives and sequence steps